License Plate Recognition Technology Combining With Super Resolution Reconstruction Of The Image Sequences

License plate recognition system refers to the computer recognition of vehiclelicense plate information in video or image, a processing system and the output of thecharacter information, the application is very extensive in daily life. However, due tothe video image causes adverse weather cloudy, rain, haze or imaging equipment itselfcaused by the fuzzy phenomenon, is the root cause of the current license platerecognition system to improve. The study combined with the image of license platerecognition sequence super resolution reconstruction is a sequence of low resolutionimages, mainly includes three parts: super resolution reconstruction, vehicle licenseplate location and character segmentation, license plate character recognitionsequence of low resolution images. The main work is as follows:1．According to the characteristics of high resolution sequence of low resolutionimages, the super-resolution reconstruction algorithm of a L1and L2hybrid model.That combines the advantages of super resolution reconstruction algorithm of L1andL2of the two paradigms that image noise can be sufficiently suppressed, details of asequence of low resolution images to highlight. The method has good robustnessunder different conditions.2．In view of the high resolution image after super resolution reconstruction ofimage sequence, location, using the method of license plate location based on thecombination of HSV color model and the method of Otsu. Firstly, the image isreconstructed from RGB space to HSV space, the color filter out irrelevantinformation for the license plate of the color space corresponding to the range. Thenthe license plate area range corresponding RGB color space images into gray imageusing Otsu method to locate the license plate area of the image to get the license platearea. Then after the positioning plate region vertical projection operation, license plateinformation to obtain a binary image in a single character, wait for the next characterrecognition.3．According to the letter of the license plate characters in seven digits positionhas the characteristics of uncertainty, using a single character library for characterrecognition process. Character recognition methods used in this paper is the gradient direction histogram and SVM combination of license plate character recognition. First,a large number of samples were collected under the normal34-character license plateof a binary image, gradient direction histogram binary image feature extraction andsupport vector machine training to get into the best classifier. Finally, after the splitbinary license plate according to the order of a single character segmentationnormalized after treatment into the classifier to identify and get the results to identifythe order of output to get the license plate characters. Experimental results show thatin contrast to the traditional license plate recognition algorithm based on featurematching and BP neural network to identify the character of a significantimprovement. The results show that by L1and L2hybrid super-resolutionreconstruction process paradigm, will identify the direction of the gradient histogrammethod and support vector machine combining the characters of the license plate hasa better recognition results.